Ueli Rutishauser: State dependent computation using coupled recurrent networks
Seminar by Ueli Rutishauser of the California Institute of Technology given to the Redwood Center for Theoretical Neuroscience at UC Berkeley on April 16, 2008.
Run time About 1 hourProducer Redwood Center for Theoretical NeuroscienceAudio/Visual sound, color
Although procedural information processing composed of conditional decisions is a hallmark of intelligent behavior, its neuronal implementation remains an open question. Physiological experiments have reported behavioral-state encoding neurons in the frontal cortices, but the organization of the neuronal circuits that could support such state-dependent processing is very poorly understood. In recent years, neuroanatomical studies have demonstrated rich inter-connections between neurons in the superficial layers of the cortex, and theoretical models have explained how recurrent connections within small populations of neurons can support co-operative competitive dynamics. We show by theoretical analysis and simulations how these circuits can embed reliable robust neuronal finite state-machines, which could support generic conditional processing in the neocortex. We demonstrate how a multi-stable neuronal network that embeds a number of states can be created very simply, by coupling two recurrent networks whose synaptic weights have been set within a range that offers soft winner-take-all (sWTA) performance. The two sWTAs have simple, homogenous locally recurrent connectivity except for a small fraction of recurrent cross-connections between them that are used to embed the required states. This coupling between the maps allows them to retain their current state after the input that elicted that state is withdrawn. A small number of 'transition neurons' implement the necessary input-driven transitions between the embedded states. We provide simple rules to systematically design and construct an arbitrary neuronal state machine composed of nearly identical recurrent maps. The significance of our finding is that it offers a method whereby the cortex could achieve a broad range of sophisticated processing by only limited specialization of the same generic neuronal circuit. joint work by Ueli Rutishauser (Caltech) and Rodney J. Douglas (INI/ETH Zurich)